This page provides an independent statistical analysis and educational interpretation of publicly reported results. ClinicalTrials.gov provides the official trial registry record. The ClinicalTrials.gov record identifies the primary endpoint and report that results were posted, but they do not provide formal statistical analyses for the primary endpoint.
1. Trial at a Glance
TWILIGHT was a completed phase 4, randomized, parallel-group, quadruple-masked trial in cardiology and interventional cardiology. The registry lists 9006 participants and evaluates ticagrelor with aspirin or alone in high-risk patients after coronary intervention.
| Feature | TWILIGHT |
|---|---|
| Trial name | TWILIGHT |
| Brief title | Ticagrelor With Aspirin or Alone in High-Risk Patients After Coronary Intervention |
| Phase | Phase 4 |
| Status | Completed |
| Therapeutic area | Cardiology |
| Conditions | Cardiovascular Disease; Interventional Cardiology |
| Allocation | Randomized |
| Design model | Parallel |
| Masking | Quadruple |
| Primary purpose | Supportive care |
| Enrollment | 9006 |
| Lead sponsor | Icahn School of Medicine at Mount Sinai |
| Sponsor type | Other |
| Primary endpoints registered | 1 |
| Primary endpoint type | Binary |
| Results posted | Yes |
| Outcome measures posted | 2 |
| ClinicalTrials.gov identifier | NCT02270242 |
2. Clinical Question
The registry describes TWILIGHT as a randomized trial of ticagrelor with aspirin or alone in high-risk patients after coronary intervention. The primary registered endpoint is the number of participants with clinically relevant bleeding, defined using Bleeding Academic Research Consortium criteria, during the 12 months after randomization.
Population
High-risk patients after coronary intervention, in the setting of cardiovascular disease and interventional cardiology.
Interventions
The registry identifies aspirin, placebo, and ticagrelor as the study interventions.
Comparator structure
The two-arm design compares ticagrelor with aspirin against ticagrelor without aspirin, represented in the safety data as Aspirin + Ticagrelor and Placebo + Ticagrelor.
Primary question
What is the frequency of first occurrence of clinically relevant BARC Type 2, 3, or 5 bleeding during the 12 months after randomization?
3. Trial Design
Placebo + Ticagrelor
- Serious adverse events: 262/3555 affected/at risk
- The registry identifies placebo and ticagrelor among the study interventions.
Aspirin + Ticagrelor
- Serious adverse events: 370/3564 affected/at risk
- The registry identifies aspirin and ticagrelor among the study interventions.
4. Timeline and Trial Status
Trial start
The registry lists July 2015 as the trial start date.
Primary completion
The registry lists July 10, 2019 as the primary completion date.
Registry status
The trial is listed as completed, with results posted and 2 outcome measures posted.
5. Primary Endpoint
| Endpoint | Registry definition | Time frame | Endpoint type |
|---|---|---|---|
| Number of Participants With BARC Type 2, 3, or 5 | Number of participants with first occurrence of clinically relevant bleeding episode, defined as Bleeding Academic Research Consortium (BARC) Types 2, 3 or 5 bleeding. BARC bleeding types range from 0 (no bleeding) to 5 (fatal bleeding). | 12 months after randomization | Binary |
The endpoint is explicitly a first occurrence endpoint. That wording is statistically important: the target quantity is whether a participant experiences at least one qualifying bleeding event during the specified follow-up period, rather than the total number of bleeding episodes.
The registry's binary classification converts each participant's follow-up experience into an outcome indicating whether the qualifying first event occurred during the defined 12-month period.
6. Statistical Methodology
The ClinicalTrials.gov record classifies the primary endpoint as binary and state that results are posted, but the ClinicalTrials.gov record contains no formal statistical analyses. Therefore, the ClinicalTrials.gov record does not identify a formal treatment-effect estimate, confidence interval, p-value, regression model, test statistic, or analysis population for the primary endpoint.
For a binary endpoint such as first occurrence of BARC Type 2, 3, or 5 bleeding by a specified time, a conventional analysis would begin with the proportion of participants experiencing the endpoint in each randomized group. A treatment comparison could then be expressed using an absolute risk difference, risk ratio, or odds ratio, depending on the prespecified statistical analysis plan.
Risk difference
The difference between the event proportions in the two randomized groups. It answers an absolute question: how much higher or lower is the observed probability of the event?
Risk ratio
The event probability in one randomized group divided by the event probability in the other. It describes relative frequency rather than an absolute difference.
Odds ratio
The odds of the event in one group relative to the odds in the other. An odds ratio is not generally interchangeable with a risk ratio, especially when the event is not rare.
Time-to-event analysis
If the timing of bleeding and censoring were central to the prespecified analysis, a time-to-event approach could use information about when the first event occurred rather than reducing follow-up solely to a fixed-period binary status.
Which of these approaches was actually prespecified for TWILIGHT's primary endpoint cannot be established from the registry-reported statistical-analysis field, because that field is empty. An educational explanation of possible methods should therefore not be mistaken for a reconstruction of the trial's formal statistical analysis.
7. Planned Analysis
Because the ClinicalTrials.gov record contains no formal statistical analyses, the appropriate statistical interpretation is to describe what would normally be measured and how a binary endpoint of this type would be compared, without asserting an unreported treatment effect.
| Analysis component | Typical approach for this endpoint | What can be concluded from the ClinicalTrials.gov record |
|---|---|---|
| Primary outcome | Compare the proportion of randomized participants with a first BARC Type 2, 3, or 5 bleeding event during 12 months. | The endpoint is binary and defined over 12 months after randomization. |
| Effect measure | Risk difference, risk ratio, odds ratio, or another prespecified measure could be reported. | No formal treatment-effect estimate is in the posted results. |
| Uncertainty | A confidence interval would normally accompany the estimated treatment effect. | No confidence interval is in the posted results. |
| Hypothesis test | A prespecified two-group test or regression-based comparison could be used. | No formal p-value or test statistic is in the posted results. |
| Follow-up | Participants would be classified according to whether the qualifying first event occurred during the specified period, with any prespecified handling of incomplete follow-up. | The endpoint time frame is 12 months after randomization. |
This is intentionally a planned-analysis explanation, not a claim about what TWILIGHT actually used in its final statistical analysis. The distinction matters because statistical methods are part of the estimand: changing the outcome definition, censoring rules, analysis population, or effect measure can change the numerical quantity being estimated.
8. Statistical Methods Explained
What makes the primary endpoint binary?
The registry defines the outcome as the number of participants with a first occurrence of BARC Type 2, 3, or 5 bleeding. Each participant can therefore be classified according to whether the qualifying endpoint occurred during the specified follow-up. A binary endpoint does not mean that the clinical event itself is simple; it means that the prespecified analysis reduces the participant-level outcome to event versus no qualifying event.
Why does "first occurrence" matter?
If one participant has several qualifying bleeding episodes, a first-occurrence endpoint does not count every episode as a separate primary endpoint event. The statistical question becomes whether and when the first qualifying event occurred. This avoids giving participants with multiple events disproportionate weight in a simple participant-level binary analysis.
Why is the 12-month time frame important?
A binary event proportion is meaningful only with a defined observation period. "Bleeding occurred" and "bleeding occurred within 12 months after randomization" are different estimands. The registry specifies the latter, so an analysis based on a different follow-up window would answer a different question.
Why is randomization important for the comparison?
Randomization is intended to make treatment assignment independent of measured and unmeasured baseline characteristics in expectation. That provides the foundation for a causal comparison of the randomized strategies, provided the subsequent analysis follows the prespecified design and appropriately handles follow-up, treatment exposure, and outcome ascertainment.
Why does quadruple masking matter?
Masking is particularly relevant when an outcome could be influenced by clinical behavior or assessment. In a quadruple-masked design, knowledge of treatment assignment is restricted across multiple roles. This can reduce opportunities for treatment knowledge to influence trial conduct, although masking cannot eliminate every possible source of bias.
What would a confidence interval add?
A confidence interval would describe statistical uncertainty around an estimated treatment effect under the chosen analysis framework. It would not describe the range of effects experienced by individual patients, and it would not establish that every patient would have the same treatment effect. Without the actual effect estimate and confidence interval, no precision statement about the primary treatment comparison can be made from the ClinicalTrials.gov record.
9. Interpreting Binary Treatment Effects
Suppose a trial reports an event proportion in each randomized group. There are several complementary ways to express the comparison. They answer different questions and should not be treated as interchangeable.
| Measure | Conceptual calculation | Interpretive question |
|---|---|---|
| Risk difference | Risk in group A − risk in group B | How many more or fewer events are observed per unit of participants? |
| Risk ratio | Risk in group A ÷ risk in group B | How many times the event probability is observed in one group relative to the other? |
| Odds ratio | Odds in group A ÷ odds in group B | How do the event odds compare between groups? |
For a clinically relevant bleeding endpoint, absolute and relative measures can lead to different impressions. A relative measure can communicate proportional differences efficiently, while a risk difference directly describes the change in event probability. The appropriate primary measure should be determined by the trial's prespecified estimand and analysis plan.
A treatment effect estimate is not meaningful in isolation from its definition. The reader needs to know the endpoint, follow-up period, analysis population, comparison direction, and effect measure before interpreting the magnitude of an observed difference.
If a formal p-value were reported, it would quantify the compatibility of the observed data with a specified null hypothesis under the chosen statistical model or test. It would not measure the size of the treatment effect, the probability that the null hypothesis is true, or the clinical importance of the result.
A confidence interval would communicate the statistical precision of the estimated effect under the specified inferential framework. A narrow interval would indicate greater precision than a wide interval, but precision and clinical importance are distinct concepts.
10. Safety Results
The ClinicalTrials.gov record provides serious adverse events by arm as affected participants divided by participants at risk. These are the safety counts available for this analysis.
| Safety measure | Placebo + Ticagrelor | Aspirin + Ticagrelor |
|---|---|---|
| Serious adverse events | 262/3555 | 370/3564 |
Placebo + Ticagrelor
Serious adverse events were reported for 262 of 3555 participants at risk.
Aspirin + Ticagrelor
Serious adverse events were reported for 370 of 3564 participants at risk.
The denominators are important. A numerator alone cannot be interpreted as a risk or rate; the number of participants at risk supplies the denominator needed to understand the frequency within each arm.
11. Limitations
The registry explicitly identifies two important limitations. First, the trial was not powered to detect differences in the risk of important but rare clinical events, including stent thrombosis and stroke. Second, the trial may not be generalizable to all patients undergoing PCI because of its high-risk criteria.
| Limitation | Statistical implication |
|---|---|
| Rare clinical events | Even a large randomized trial may have limited precision for uncommon outcomes. A low event count can produce wide uncertainty around a treatment comparison. |
| High-risk eligibility criteria | The enrolled population may differ systematically from the broader population of patients undergoing PCI. External validity therefore requires attention to the trial's eligibility criteria. |
| Missing formal statistical-analysis details in the ClinicalTrials.gov record | The ClinicalTrials.gov record does not establish the exact treatment-effect measure, inferential test, confidence-interval method, missing-data strategy, or analysis population used for the primary endpoint. |
| Binary endpoint | A fixed-period binary endpoint summarizes whether an event occurred, but may discard information about the exact timing of the event unless the analysis explicitly incorporates time-to-event information. |
These limitations do not invalidate the randomized design. Instead, they define what can and cannot be learned from the available evidence. In particular, lack of power for rare outcomes is a precision limitation, while restrictive eligibility criteria primarily affect generalizability.
12. Important Statistical Interpretation Issues
Randomization does not guarantee identical groups
Randomization balances prognostic factors in expectation, not necessarily in every realized sample. With 9006 enrolled participants, random variation in baseline characteristics can still occur. The principal statistical value of randomization is the validity of the treatment-assignment mechanism, not numerical identity between groups.
Masking reduces one class of bias but does not solve every problem
Quadruple masking can reduce the influence of treatment knowledge on trial conduct and assessment. It does not automatically eliminate selection bias, incomplete follow-up, measurement error, nonadherence, or limitations in generalizability.
A binary endpoint is not the same as a rate
The primary endpoint counts participants with a first qualifying bleeding episode over a defined 12-month period. A rate would additionally incorporate person-time at risk. Confusing a proportion with an incidence rate can lead to incorrect interpretation, especially if follow-up differs substantially among participants.
Censoring matters if time-to-event methods are used
If the statistical analysis incorporates the timing of bleeding, participants without a qualifying event by their last known follow-up contribute information up to that point rather than necessarily being treated as having complete 12-month observation. The correct censoring rules are part of the prespecified analysis and are not reported in the provided statistical-analysis field.
Clinical importance and statistical significance are different
Even if a formal comparison produces a small p-value, that does not determine whether the magnitude of the difference is clinically important. Conversely, an important-looking point estimate can be imprecise when event counts are small. Effect size, uncertainty, and clinical context should be interpreted together.
13. Why This Trial Matters Statistically
TWILIGHT is a useful teaching case because the primary outcome is a clinically meaningful binary endpoint embedded within a randomized, parallel-group, quadruple-masked trial. The statistical challenge is not simply to count bleeding events. It is to define exactly which participants count as events, over what time frame, and relative to which randomized treatment strategy.
| Concept | How it appears in TWILIGHT |
|---|---|
| Randomization | The registry identifies the allocation as randomized. |
| Parallel design | The design model is listed as parallel with 2 arms. |
| Quadruple masking | The registry lists masking as quadruple. |
| Binary endpoint | The primary endpoint is classified as binary. |
| First-event analysis | The primary endpoint counts first occurrence of clinically relevant bleeding. |
| Fixed follow-up window | The registered time frame is 12 months after randomization. |
| Composite clinical definition | The qualifying endpoint includes BARC Types 2, 3, or 5 bleeding. |
| Safety analysis | Serious adverse events are reported by arm with affected/at-risk counts. |
| External validity | The registry cautions that high-risk criteria may limit generalizability to all patients undergoing PCI. |
| Rare-event precision | The registry notes limited power for important but rare events such as stent thrombosis and stroke. |
14. Clinical Question Versus Statistical Estimand
A useful distinction in modern clinical-trial statistics is between the broad clinical question and the precise statistical quantity being estimated. "Does ticagrelor with aspirin or alone change bleeding?" is clinically understandable but statistically incomplete.
Population
High-risk patients after coronary intervention represented by the randomized trial population.
Treatment comparison
Ticagrelor with aspirin versus ticagrelor alone, represented in the ClinicalTrials.gov record as Aspirin + Ticagrelor versus Placebo + Ticagrelor.
Outcome
First occurrence of clinically relevant BARC Type 2, 3, or 5 bleeding.
Time horizon
12 months after randomization.
Making those elements explicit prevents a common statistical error: interpreting a result from one endpoint or time horizon as though it answers a broader question about all bleeding, all cardiovascular outcomes, or all patients undergoing coronary intervention.
15. What the Registry Data Do and Do Not Establish
| Question | Supported by the ClinicalTrials.gov record? | Reason |
|---|---|---|
| Was TWILIGHT randomized? | Yes | Allocation is listed as RANDOMIZED. |
| Was the design parallel? | Yes | Design model is listed as PARALLEL. |
| Was the trial quadruple-masked? | Yes | Masking is listed as QUADRUPLE. |
| Was the primary endpoint binary? | Yes | Primary endpoint type is listed as Binary. |
| What was the primary endpoint? | Yes | The registry definition is reported in full. |
| What was its time frame? | Yes | 12 months after randomization. |
| How many participants were enrolled? | Yes | Enrollment is 9006. |
| Were results posted? | Yes | The record states Results Posted: Yes. |
| What was the primary hazard ratio? | No | No statistical-analysis estimate is reported. |
| What was the primary confidence interval? | No | No statistical-analysis confidence interval is reported. |
| What was the primary p-value? | No | No statistical-analysis p-value is reported. |
| What exact inferential model was used? | No | No formal statistical analyses were posted. |
This distinction is central to responsible statistical reporting. A posted result and a documented formal statistical analysis are not synonymous. Where the available record does not identify the formal comparison, an educational page should preserve that uncertainty rather than reverse-engineering an unverified analysis.
16. Serious Adverse Events: Statistical Perspective
The serious-adverse-event data illustrate why denominators and comparison measures matter. The registry-reported counts are 262/3555 for Placebo + Ticagrelor and 370/3564 for Aspirin + Ticagrelor.
For example, the registry supplies both numerator and denominator for each arm. A formal comparison would require an explicitly defined statistical method and inferential framework; none is provided in the posted results.
It is also important not to assume that a serious adverse event is equivalent to the primary BARC bleeding endpoint. Serious adverse events encompass a safety category, whereas the primary endpoint has a specific clinical definition based on BARC Types 2, 3, or 5 bleeding.
17. Trial History
Study begins
The registry lists July 2015 as the start of TWILIGHT.
Primary completion
The registry lists July 10, 2019 as the primary completion date.
Results posted
The registry marks the study as completed and reports that results have been posted, with 2 outcome measures posted.
18. Statistical Interpretation: What Would Make a Result Informative?
The primary endpoint is not simply "bleeding." It is the number of participants with first occurrence of clinically relevant bleeding episode defined as BARC Types 2, 3, or 5, during 12 months after randomization.
A useful treatment comparison would state the event frequency in both randomized groups and provide a clearly defined effect measure such as a risk difference, risk ratio, or odds ratio, according to the prespecified analysis.
The point estimate should be accompanied by an appropriate confidence interval. Without that interval, the reader cannot judge the statistical precision of the estimated treatment effect.
A statistically detectable difference is not automatically clinically important, and a clinically important point estimate may remain uncertain when event counts are limited. Both magnitude and precision are necessary for interpretation.
Randomized efficacy comparisons and exposure-based safety summaries can use different analysis populations. The ClinicalTrials.gov record provides serious adverse events by arm but does not specify a complete formal analysis-population definition for the primary endpoint.
19. Limitations of the Available Statistical Record
The primary statistical limitation of the ClinicalTrials.gov record is not a flaw in the trial design; it is a limitation in the level of statistical detail provided in the extracted record. No formal statistical analyses were posted despite the record stating that results are posted.
- No formal effect estimate: the ClinicalTrials.gov record does not provide a treatment-effect estimate for the primary BARC endpoint.
- No confidence interval: the ClinicalTrials.gov record does not provide an uncertainty interval for the primary comparison.
- No p-value: the ClinicalTrials.gov record does not identify a formal hypothesis-test result.
- No named model: the ClinicalTrials.gov record does not identify whether the primary comparison used a simple proportion test, regression model, time-to-event method, or another approach.
- No formal missing-data strategy: the ClinicalTrials.gov record does not identify how incomplete follow-up was handled for the primary analysis.
- No multiplicity strategy reported: the provided record does not establish an alpha-allocation strategy for the primary and secondary outcomes.
- No interim-analysis information reported: the provided record does not establish whether or how interim statistical monitoring was performed.
- No Bayesian methodology reported: the provided record does not identify Bayesian modeling for the primary endpoint.
These omissions are reasons to avoid overinterpretation, not reasons to fill the gaps with assumptions. A rigorous statistical page should distinguish documented trial facts from methods that are merely typical for a particular endpoint.
20. Why Randomization and Blinding Matter Here
Randomization
Random allocation provides the framework for comparing the two treatment strategies without deliberately assigning treatment according to baseline prognosis.
Quadruple masking
Masking can reduce the opportunity for knowledge of treatment assignment to influence participant behavior, clinical care, outcome assessment, or other trial processes.
Parallel groups
Each randomized participant contributes to one assigned strategy rather than moving through both randomized treatment groups in a crossover comparison.
Endpoint standardization
The use of BARC Types 2, 3, or 5 creates a prespecified clinical definition rather than leaving "clinically relevant bleeding" entirely to subjective interpretation.
These design elements work together. Randomization primarily supports causal comparison, masking can reduce ascertainment and behavioral biases, and a standardized endpoint improves consistency in determining what counts as an event. None of these features, individually or collectively, supplies the missing numerical treatment-effect estimate.
21. Related Statistical Tutorials
Learn more about the methods used in this trial:
22. Related Statistical Calculators
23. Sources
- ClinicalTrials.gov: NCT02270242 — TWILIGHT.
- Linked publication: PubMed PMID 39889917.
- Linked publication: PubMed PMID 38458371.
- Linked publication: PubMed PMID 37806185.
- Linked publication: PubMed PMID 37587580.
- Linked publication: PubMed PMID 37094668.
Continue through the Clinical Biostats statistical pathway
Use the trial's design concepts to explore broader methods for randomized trials, binary endpoints, masking, and statistical interpretation.
24. Record Summary
TWILIGHT is a completed phase 4, randomized, parallel-group, quadruple-masked trial with 9006 enrolled participants. Its registered primary endpoint is a binary measure of the first occurrence of clinically relevant bleeding defined as BARC Type 2, 3, or 5 bleeding during the 12 months after randomization. The ClinicalTrials.gov record also report serious adverse events of 262/3555 in the Placebo + Ticagrelor arm and 370/3564 in the Aspirin + Ticagrelor arm.
The statistical lesson is equally important: a rigorous clinical-trial analysis requires more than a posted result. It requires a clearly defined estimand, analysis population, effect measure, uncertainty interval, and inferential procedure.